臺大管理論叢 NTU Management Review VOL.28 NO.3

Use of Implicit User Feedback to Support Semantics-Based Personalized Document Recommendation 88 action. User preferences are gathered from streaming data including the click log, viewing times and browsing paths. Since explicit elicitation requires deliberate user action, it is considered to be less hospitable and more costly than implicit elicitation. Prior research has also suggested that these two elicitation techniques perform about the same in user- based collaborative filter recommender systems (Jawaheer, Szomszor, and Kostkova, 2010). To overcome the shortcomings of explicit elicitation, implicit elicitation has recently been adopted to reduce user effort and enhance the interactive experience. For example, PRemiSE, a novel personalized news recommendation framework, considers the opinions of potential social experts to make recommendations for the focal user (Lin, Xie, Li, and Li, 2014). The opinions of potential influencers on virtual social networks are used as auxiliary resources for recommendations. Lin et al. (2014) combined the semantic meanings of news stories and the structure of implicit user feedback to generate news story recommendations. To address the problem of data sparsity, Zhang, Wang, and Yi (2014) proposed the innovative Adaptive Recommendation Algorithm (ARA) approach, based on a small-world implicit trust network. Yu, Ma, Hsu, and Han (2014) applied an entity recommender system to make personalized movie recommendations based on user click logs, which can be easily collected from a widely used commercial search engine, eliminating the need for users to respond to a questionnaire. Bauer and Nanopoulos (2014) proposed a quantitative implicit customer feedback mechanism based on users’ sale and play records to support a recommender system using matrix factorization. The results of an experimental evaluation based on three real world datasets suggest the effectiveness of their approach. Overall, the implicit feedback technique has notable advantages over the explicit feedback approach in that it allows for the creation of objective user profiles without extra user effort, and its performance is comparable with that of the explicit feedback approach. 3. Design of The Implicit-Feedback-Based Concept-Expansion Document Recommendation (ICE) Technique To address the inherent difficulties in acquiring user preference feedback, this study proposes the ICE technique which analyzes browsing behavior to identify documents in which the focal user may be interested. We also adopt the Spreading Activation Model (SAM) to discover concepts relevant to the users’ preferred documents in order to enhance